If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below. |
|
|
Thread Tools | Display Modes |
#21
|
|||
|
|||
natric wrote:
reply inline Hummm... Just reading your last post now. Well, I agree with you that the ram timings has to be considerated deeper than I've done (and more since it's actually not in manual mode), but remembering my final objective is to push up global performances : then, thinking twice for every setting. During last night and as said in my last post (before yours), I've ran Prime95 two times with Torture Test's Small FFTs option, trying to see if cpu side is right enough before to closer considerate the RAM aspect. Here are the results : 1) With VCORE @ 1.75V : error after 5 hours (under our 6h threshold) 2) With VCORE @ 1.775V : still running since 8 hours without error and temp. is 55°C (around [36°C, 41°C] in usual daily use)... Then, from this step, I'm thinking VCORE is high enough (and remembering I'm not in a high cooling environment) and will considerate the RAM side details and other stuff in the mobo. Well, this push toward some question. Leave the Vcore at 1.775V if the system is stable. 1) About RAM voltage : originally @ 2.6V, now @ 2.7V, is there some risk to go to 2.8V (not any smaller step) and does-it present some interest ? Try not to increase RAM voltage past 2.7V. In most cases, you can leave it at 2.6V, when the system is stable. 2) About timings : actually as 8-4-4-2.0, first you preconised 11-3-3- 3.0 (but it did generated beep codes) and 11-4-4-3.0 in your last post, then I suppose your message is to say that current CAS is too fast, that (as I've read) the "11" is often seen for nForce2 case and two last middle values are maybe to be privilegiated from the CAS. On my side, I'm starting to search info about this RAM timings point and, for example have found url like this one http://www.frazpc.pl/artykuly_wyswietl/212_en that seems to be very instructive... But, on your side, do you know some step-by-step guide or valuable urls about this specific subject ? What's your own methodology ? Using Corsair PC3200, I've found 11-3-3-2.5 timings to produce the best benchmarks at 200MHz, with 2.6V DDR voltage. The Corsair is capable of 6-3-3-2.5 timings. It was all experimentation. 3) A kind of subsidiary question : what's about chipset voltage ? Does- it has some sense to consider-it ? How ? Chipset voltage is nominally 1.5V. I leave it at default. An increase is not normally called for. |
#22
|
|||
|
|||
natric wrote:
[reply inline; snipped where necessary] Prime95's Blend Test Torture has been right during 9h30 and I've stopped it manually (since some sensible work to do in the pc). Current stable state is : CPU @ 11*200MHz on 1.775V, RAM on 2.7V and timings as 8-4-4-2.0, Chipset and AGP voltage stay unchanged. You can try lowering RAM voltage to 2.6V if the system is otherwise stable. Now, when I'll have time, I'll take some try around different RAM timings to see benchmark incidence. As promised, I've made some trial with ram timings and here are my representative tests. All under a "charged" Win2K without any special preparation (ie. a lot of things in systray, including something like TuneUp MemOptimizer which may slow down instant perf occasionnaly) and knowing the nf7-s is working in DualBand with 2*256MB of noname with Samsung chips. Here are some representative results : timings / aida32 read & write / sandra2003 int & float 8-4-4-2.0 (SPD one) / 3042 & 1157 / 3006MB/s & 2705MB/s 7-3-3-2.0 (Turbo one) / 3070 & 1175 / 3036 & 2841 (best sandra float) 10-3-3-2.0 / 3077 & 1178 / 3037 & 2758 (globally the best) Using 11-3-3-2.5 timings at 200MHz, 2.6V, I get: AIDA32 ver3.93 2971/1103 (read/write) SANDRA 2004.2.9.104 2954/2720 (integer/float) It looks like the lower CAS of 2.0 helps. I would keep the 10-3-3-2 timings. From this point, I'm testing the stability in both, starting with the BIOS preset one : 7-3-3-2.0 = A Prime95 Blend Test Torture (this time, MemOptimizer being closed to avoid any interference) has been ran during 10h without error and I've stopped it manually. About temperature : it was 38°C before launching and has gone up to 55°C. Then, I can considerate this config is honnestly stable. Next steps will be to do this same Prime95 test with the 10-3-3-2.0... And, maybe (but require lot of time), to try exhaustively all the possible combinations (especially all the ones with the 10,11,12 tRAS as recommended for the nForce2) :-p You could, but it's probably not necessary. And this, without to talk about another way which would be to increase FSB again (but, surely would require to downgrade CAS). Without to talk about the fact, I should, certainly, be inspired to decrease VCORE and RAM voltage, for the purpose to find the minimum stable one in final state. Without to talk about the possible importance around chipset voltage. Without to talk about the pertinence of my, finally, reasonable cooling elements. If you want to lower a voltage, lower only one, by the smallest amount possible. Test. In this way you will know which settings to revert to if it becomes unstable. However, a real and strong methodology around ram timings choices remains welcome if you have one !!! Let me (us) know Using PC3200, I start at 11-3-3-3 or 11-4-4-3 with V RAM at 2.6V and 1.65V Vcore, both memory and FSB at 200MHz. If the system is stable, I try slightly more aggressive RAM timings; one parameter at a time. The very first change I make if the system is unstable is more V core. To find errors more quickly, I find SuperPi to be a good tool. http://pw1.netcom.com/~hjsmith/Pi/Super_Pi.html The longest test, 32 million digits, can be accomplished in ~40 minutes. Any errors usually point to insufficient Vcore or too aggressive RAM timings. Prime95 can then be used to check for long term stability. |
#24
|
|||
|
|||
natric wrote:
Ok, Shaw, thanks for this url. Just downloaded it and will go through process with it first and Prime95 later. However, I agree with the 10-3-3-2.0 adoption at this step (will, maybe, just take a quick try with 11-3-3-2) and will try to see if ram volatage decreasment will have impact on stability. Thanks again for your help Have a good day My pleasure Natric. |
#25
|
|||
|
|||
[Posted to alt.comp.mainboards.abit and copy sent to the cited author]
In article , says... [Posted to alt.comp.mainboards.abit and copy sent to the cited author] In article uZ3pc.476223$oR5.119837@pd7tw3no, says... To find errors more quickly, I find SuperPi to be a good tool. http://pw1.netcom.com/~hjsmith/Pi/Super_Pi.html The longest test, 32 million digits, can be accomplished in ~40 minutes. Any errors usually point to insufficient Vcore or too aggressive RAM timings. Prime95 can then be used to check for long term stability. Ok, Shaw, thanks for this url. Just downloaded it and will go through process with it first and Prime95 later. However, I agree with the 10-3-3-2.0 adoption at this step (will, maybe, just take a quick try with 11-3-3-2) and will try to see if ram volatage decreasment will have impact on stability. Thanks again for your help Have a good day As promised, final results : As said in last post, I've tried to decrease voltage and see if stability if always at the rendez-vous : reply is "no" ! Here are the details (@11*200 & RAM timings as 10-3-3-2.0): 1) VCORE 1.775, RAM 2.7 = 41°C idle, SuperPI 32M is OK, Prime95 Blend Torture Test runs during 11h30 (55°C max.) before error :-)) 2) RAM decreased to 2.6V = Prime95 BTT runs during 12mn before error :- ( 3) RAM raised to 2.7V again, but VCORE decreased to 1.75V = 40°C idle, SuperPI 32M is OK, Prime95 BTT runs during 1h15 before error :-( 4) I return to VCORE 1.775V and RAM to 2.7V = 44°C (today is a hot day), Prime95 BTT runs during 6h12 (59°C max.) before error :-) Then, we can say, it's very voltage and temperature dependent. I have to stay at VCORE 1.775V and RAM at 2.7V. End of the o/c |
#26
|
|||
|
|||
natric wrote:
As promised, final results : As said in last post, I've tried to decrease voltage and see if stability if always at the rendez-vous : reply is "no" ! Here are the details (@11*200 & RAM timings as 10-3-3-2.0): 1) VCORE 1.775, RAM 2.7 = 41°C idle, SuperPI 32M is OK, Prime95 Blend Torture Test runs during 11h30 (55°C max.) before error :-)) 2) RAM decreased to 2.6V = Prime95 BTT runs during 12mn before error :- ( 3) RAM raised to 2.7V again, but VCORE decreased to 1.75V = 40°C idle, SuperPI 32M is OK, Prime95 BTT runs during 1h15 before error :-( 4) I return to VCORE 1.775V and RAM to 2.7V = 44°C (today is a hot day), Prime95 BTT runs during 6h12 (59°C max.) before error :-) Then, we can say, it's very voltage and temperature dependent. I have to stay at VCORE 1.775V and RAM at 2.7V. End of the o/c Tout est bien qui finit bien. :-) |
#27
|
|||
|
|||
|
Thread Tools | |
Display Modes | |
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
What would be the quietest 80mm fan with with the highest CFM ? | We Live For The One We Die For The One | General | 0 | May 1st 04 09:24 AM |
What would be the quietest 80mm fan with with the highest CFM ? | We Live For The One We Die For The One | Homebuilt PC's | 0 | May 1st 04 09:23 AM |
My Highest Temps. | Dennis E Strausser Jr | Overclocking | 3 | March 26th 04 07:32 AM |
P4c-3.2Ghz - what's the highest speed record so far? | Vincent Poy | Overclocking | 5 | August 11th 03 05:37 PM |
highest fsb for kt400 | Dakota650r | Overclocking AMD Processors | 0 | August 10th 03 04:52 AM |